| a40285c | Thiago Crepaldi | 22 January 2018, 20:34:25 UTC | Rename default Anaconda python environment Depending on the CNTK setup steps followed, either cntk-py<version> or <cntkdev-py<version> is installed. cntk-py<version> will be adopted as the default from now on. | 23 January 2018, 18:43:51 UTC |
| e883520 | Thiago Crepaldi | 23 January 2018, 01:08:35 UTC | Remove redefinition of CNTK_COMPONENT_VERSION Math.vcsproj inherits this macro from CNTK.Common.props already | 23 January 2018, 18:28:53 UTC |
| 3cf3af5 | KeDengMS | 23 January 2018, 00:58:56 UTC | CNTK support for CUDA 9 CNTK now supports CUDA 9/cuDNN 7. This requires an update to build environment to Ubuntu 16/GCC 5 for Linux, and Visual Studio 2017/VCTools 14.11 for Windows. With CUDA 9, CNTK also added a preview for 16-bit floating point (a.k.a FP16) computation. Please check out the example of FP16 in ResNet50 at /Examples/Image/Classification/ResNet/Python/TrainResNet_ImageNet_Distributed.py Notes on FP16 preview: * FP16 implementation on CPU is not optimized, and it's not supposed to be used in CPU inference directly. User needs to convert the model to 32-bit floating point before running on CPU. * Loss/Criterion for FP16 training needs to be 32bit for accumulation without overflow, using cast function. Please check the example above. * Readers do not have FP16 output unless using numpy to feed data, cast from FP32 to FP16 is needed. Please check the example above. * FP16 gradient aggregation is currently only implemented on GPU using NCCL2. Distributed training with FP16 with MPI is not supported. * FP16 math is a subset of current FP32 implementation. Some model may get Feature Not Implemented exception using FP16. * FP16 is currently not supported in BrainScript. Please use Python for FP16. To setup build and runtime environment on Windows: * Install [Visual Studio 2017](https://www.visualstudio.com/downloads/) with following workloads and components. From command line (use Community version installer as example): vs_community.exe --add Microsoft.VisualStudio.Workload.NativeDesktop --add Microsoft.VisualStudio.Workload.ManagedDesktop --add Microsoft.VisualStudio.Workload.Universal --add Microsoft.Component.PythonTools --add Microsoft.VisualStudio.Component.VC.Tools.14.11 * Install [NVidia CUDA 9](https://developer.nvidia.com/cuda-90-download-archive?target_os=Windows&target_arch=x86_64) * From PowerShell, run: /Tools/devInstall/Windows/DevInstall.ps1 * Start VCTools 14.11 command line, run: cmd /k "%VS2017INSTALLDIR%\VC\Auxiliary\Build\vcvarsall.bat" x64 --vcvars_ver=14.11 * Open /CNTK.sln from the VCTools 14.11 command line. Note that starting CNTK.sln other than VCTools 14.11 command line, would causes CUDA 9 [build error](https://developercommunity.visualstudio.com/content/problem/163758/vs-2017-155-doesnt-support-cuda-9.html). To setup build and runtime environment on Linux using docker, please build Unbuntu 16.04 docker image using Dockerfiles /Tools/docker. For other Linux systems, please refer to the Dockerfiles to setup dependent libraries for CNTK. | 23 January 2018, 00:58:56 UTC |
| 3765da9 | Manik Jindal | 20 January 2018, 01:59:03 UTC | Support CNTK installation in VirtualEnv CNTK installation using pip place dlls in the root folder of the python environment. Unlike Conda, VirtualEnv doesn't add this folder in library search path and hence 'import CNTK' is broken in VirtualEnv. Fix: Place CNTK libraries in site-packages/CNTK/lib and add the same to library search path. | 20 January 2018, 01:59:03 UTC |
| 997d3a1 | Thiago Crepaldi | 17 January 2018, 20:02:59 UTC | Replace CntkCsAssemblyVersion by CntkComponentVersion on VS projects | 19 January 2018, 22:36:34 UTC |
| f36c73c | Project Philly | 19 January 2018, 19:41:04 UTC | Integrate sptiwari/find_by_uid4 into master | 19 January 2018, 19:41:04 UTC |
| 4836687 | Spandan Tiwari | 17 January 2018, 23:10:23 UTC | Fixing bug in ONNX pooling node serialization. | 17 January 2018, 23:10:23 UTC |
| f07a98d | Spandan Tiwari | 17 January 2018, 22:35:23 UTC | Adding find_by_uid to logging API. | 17 January 2018, 22:35:23 UTC |
| 8066621 | Eldar Akchurin | 03 January 2018, 16:40:52 UTC | Fixing BPTT for case when the minibatch size changes | 17 January 2018, 01:20:23 UTC |
| c3d01e4 | Eldar Akchurin | 03 January 2018, 14:34:32 UTC | Enabling evaluation on GPU | 17 January 2018, 01:20:23 UTC |
| 5bdaed7 | Project Philly | 16 January 2018, 23:27:02 UTC | Integrate yuqtang/TimesOnFreeAxes into master | 16 January 2018, 23:27:02 UTC |
| 10d7130 | Project Philly | 16 January 2018, 22:39:23 UTC | Integrate vadimma/LatticeSerializer into master | 16 January 2018, 22:39:23 UTC |
| 8b6af80 | Vadim Mazalov | 16 January 2018, 03:41:23 UTC | Update the max_sequence_length | 16 January 2018, 03:41:23 UTC |
| 05cbd0a | Vadim Mazalov | 16 January 2018, 00:40:06 UTC | Remove too much logging | 16 January 2018, 00:40:06 UTC |
| e6dc1a4 | Vadim Mazalov | 15 January 2018, 23:22:11 UTC | Refactor lattice sequence initialization' | 15 January 2018, 23:22:11 UTC |
| 77c95a4 | Vadim Mazalov | 15 January 2018, 08:08:02 UTC | Trim lattice TOC line entries | 15 January 2018, 08:08:02 UTC |
| d929ecd | Vadim Mazalov | 15 January 2018, 07:45:18 UTC | Add logs for lattice index builder | 15 January 2018, 07:45:18 UTC |
| 1a81d41 | KeDengMS | 15 January 2018, 07:32:12 UTC | Fix batch matmul test failures | 15 January 2018, 07:32:12 UTC |
| d26f547 | Vadim Mazalov | 15 January 2018, 04:57:11 UTC | debug large lattice entries | 15 January 2018, 04:57:11 UTC |
| 1fe17f8 | Vadim Mazalov | 15 January 2018, 01:13:02 UTC | Add some debugging statements | 15 January 2018, 01:13:02 UTC |
| f79324c | Vadim Mazalov | 15 January 2018, 00:17:15 UTC | Ensure the lattice serializer handles several chunks in a file | 15 January 2018, 00:17:15 UTC |
| ea6834b | Vadim Mazalov | 14 January 2018, 17:57:29 UTC | Restrict the size of lattice | 14 January 2018, 17:57:29 UTC |
| 5962f71 | Vadim Mazalov | 14 January 2018, 06:54:37 UTC | Introduce the maxSequenceSize config to the HTK feature reader | 14 January 2018, 06:54:37 UTC |
| 7d67daf | Vadim Mazalov | 14 January 2018, 00:16:40 UTC | Enable max seq size config in HTK reader | 14 January 2018, 00:16:40 UTC |
| 15e705d | Chengji Yao | 08 January 2018, 07:30:53 UTC | add batch matmul | 13 January 2018, 02:41:34 UTC |
| 01c22ad | Vadim Mazalov | 13 January 2018, 01:38:02 UTC | Change the docstring | 13 January 2018, 01:38:02 UTC |
| bcbf95b | Vadim Mazalov | 12 January 2018, 20:07:47 UTC | Some fixes after rebase | 12 January 2018, 20:07:47 UTC |
| f3e9993 | Vadim Mazalov | 12 January 2018, 18:46:55 UTC | Remove main from the lattice test | 12 January 2018, 19:06:41 UTC |
| 2cac661 | Vadim Mazalov | 11 January 2018, 17:49:04 UTC | Update the python test | 12 January 2018, 19:06:41 UTC |
| 44f21e7 | Vadim Mazalov | 11 January 2018, 03:44:44 UTC | Update python test | 12 January 2018, 19:06:40 UTC |
| eb88155 | Vadim Mazalov | 10 January 2018, 21:56:10 UTC | Include the python test and bump up the model version | 12 January 2018, 19:06:40 UTC |
| 0a6a429 | Vadim Mazalov | 10 January 2018, 21:00:17 UTC | Update baselines to ensure they match | 12 January 2018, 19:03:24 UTC |
| 9c70a43 | Vadim Mazalov | 10 January 2018, 20:57:55 UTC | Python lattice reader test | 12 January 2018, 19:03:23 UTC |
| 5bb82fa | Vadim Mazalov | 10 January 2018, 08:04:27 UTC | Expose lattice reader to python | 12 January 2018, 19:03:23 UTC |
| 697793b | Vadim Mazalov | 09 January 2018, 23:07:36 UTC | Ensure the e2e tests between old and new readers match | 12 January 2018, 19:03:23 UTC |
| bf673b8 | Vadim Mazalov | 09 January 2018, 03:44:18 UTC | Disable parallel lattice constructoin | 12 January 2018, 19:03:23 UTC |
| 25d79f1 | Vadim Mazalov | 08 January 2018, 23:31:49 UTC | Update the SE with new reader test | 12 January 2018, 19:03:23 UTC |
| 73f2315 | Vadim Mazalov | 08 January 2018, 22:29:35 UTC | Ensure the clean up builds | 12 January 2018, 19:03:22 UTC |
| e57bac1 | Vadim Mazalov | 08 January 2018, 20:40:38 UTC | Further lattice reader cleanup | 12 January 2018, 19:03:22 UTC |
| 5489a4e | Vadim Mazalov | 08 January 2018, 04:25:19 UTC | Change the old lattice reader config back | 12 January 2018, 18:59:43 UTC |
| 9f1d5e2 | Vadim Mazalov | 08 January 2018, 03:59:30 UTC | Update old lattice reader test to 1 utterances | 12 January 2018, 18:59:43 UTC |
| aaaa939 | Vadim Mazalov | 08 January 2018, 03:51:33 UTC | Expose latticedeserializer in python | 12 January 2018, 18:59:43 UTC |
| ce3ff8c | Vadim Mazalov | 06 January 2018, 06:26:32 UTC | Remove baselines for update | 12 January 2018, 18:59:42 UTC |
| 75f3e43 | Vadim Mazalov | 06 January 2018, 06:26:10 UTC | Ensure lattice sequence owns the buffer | 12 January 2018, 18:59:42 UTC |
| 6f9185f | Vadim Mazalov | 05 January 2018, 20:52:24 UTC | Revert changes of the HTKMLFreader | 12 January 2018, 18:59:42 UTC |
| e230130 | Vadim Mazalov | 05 January 2018, 20:41:54 UTC | Some clean of the lattice reader change | 12 January 2018, 18:59:42 UTC |
| 8c60dca | Vadim Mazalov | 05 January 2018, 01:11:31 UTC | Ensure the new reader output is identical to the old reader | 12 January 2018, 18:59:42 UTC |
| 1d04199 | Eldar Akchurin | 03 January 2018, 12:48:47 UTC | Forcibly moving allocation of lattices to CPU | 12 January 2018, 18:59:41 UTC |
| 733a4e7 | Eldar Akchurin | 03 January 2018, 09:40:05 UTC | Fixing mbsize | 12 January 2018, 18:59:41 UTC |
| 6e39b34 | Vadim Mazalov | 03 January 2018, 08:14:30 UTC | Clean up boundary code | 12 January 2018, 18:59:41 UTC |
| 0ea1eaf | Vadim Mazalov | 01 January 2018, 16:26:03 UTC | Remove current dir logging | 12 January 2018, 18:59:41 UTC |
| d4e71b8 | Vadim Mazalov | 01 January 2018, 02:01:53 UTC | Return the SequenceTraining baselines | 12 January 2018, 18:59:41 UTC |
| 0adc7be | Vadim Mazalov | 31 December 2017, 20:11:59 UTC | Update e2e test baselines | 12 January 2018, 18:59:41 UTC |
| a5b78ee | Vadim Mazalov | 31 December 2017, 16:55:10 UTC | Some more debugging | 12 January 2018, 18:59:40 UTC |
| 8912cbf | Vadim Mazalov | 31 December 2017, 07:28:58 UTC | Some debugging of failed linux test | 12 January 2018, 18:59:40 UTC |
| fcb3a2f | Vadim Mazalov | 31 December 2017, 02:46:21 UTC | Add some logs for the lattice reader' | 12 January 2018, 18:59:40 UTC |
| 4135324 | Vadim Mazalov | 30 December 2017, 20:13:42 UTC | Add check if lattice index file exists | 12 January 2018, 18:59:40 UTC |
| bd783aa | Vadim Mazalov | 30 December 2017, 17:01:03 UTC | Debug statement for the Linux test | 12 January 2018, 18:59:40 UTC |
| c848303 | Vadim Mazalov | 29 December 2017, 23:52:51 UTC | Update E2E test for lattice reader | 12 January 2018, 18:59:40 UTC |
| 646cf60 | Vadim Mazalov | 27 December 2017, 16:33:58 UTC | Revert pre-compute break statement | 12 January 2018, 18:59:39 UTC |
| 13206a4 | Vadim Mazalov | 27 December 2017, 03:31:56 UTC | Add lattice index builder to makefile | 12 January 2018, 18:59:39 UTC |
| 4b975a0 | Vadim Mazalov | 26 December 2017, 23:13:35 UTC | Explicitly reference SequenceWithSoftmax node methods | 12 January 2018, 18:59:39 UTC |
| 554610e | Vadim Mazalov | 26 December 2017, 17:32:32 UTC | Some gcc fixes | 12 January 2018, 18:59:39 UTC |
| 9565deb | Vadim Mazalov | 26 December 2017, 16:58:39 UTC | Remove tabs | 12 January 2018, 18:59:39 UTC |
| d38c060 | Vadim Mazalov | 25 December 2017, 18:33:59 UTC | Add new lattice reader test | 12 January 2018, 18:59:39 UTC |
| 60c2996 | Vadim Mazalov | 25 December 2017, 06:55:30 UTC | Some optimization of the binary reader | 12 January 2018, 18:59:38 UTC |
| 0750688 | Vadim Mazalov | 22 December 2017, 07:44:39 UTC | some debugging statements | 12 January 2018, 18:59:38 UTC |
| 36315e7 | Vadim Mazalov | 21 December 2017, 07:20:24 UTC | Some asserts and prints | 12 January 2018, 18:59:38 UTC |
| 9041489 | Eldar Akchurin | 20 December 2017, 15:39:38 UTC | Fixing non existing sequence | 12 January 2018, 18:59:38 UTC |
| 6fc83ca | Vadim Mazalov | 20 December 2017, 00:08:40 UTC | Ensure lattice sequences matche the label sequences | 12 January 2018, 18:59:38 UTC |
| 8ab9d94 | Vadim Mazalov | 19 December 2017, 23:42:05 UTC | Fix indexing of lattice sequences | 12 January 2018, 18:59:38 UTC |
| 815e3fc | Eldar Akchurin | 19 December 2017, 14:12:52 UTC | Fixing definesMbSize for HTK deserializer | 12 January 2018, 18:59:37 UTC |
| cf28412 | Vadim Mazalov | 19 December 2017, 01:06:56 UTC | update preferred device for lattice | 12 January 2018, 18:59:37 UTC |
| f5edf14 | Vadim Mazalov | 18 December 2017, 21:47:19 UTC | Some cleanup of the SE with lattice node | 12 January 2018, 18:59:37 UTC |
| a01da0c | Vadim Mazalov | 18 December 2017, 17:18:00 UTC | Fix lattice index for last sequence | 12 January 2018, 18:59:37 UTC |
| 3dba770 | Vadim Mazalov | 18 December 2017, 08:05:30 UTC | Testing parallel sequences | 12 January 2018, 18:59:37 UTC |
| 8c8560b | Vadim Mazalov | 17 December 2017, 20:28:56 UTC | Initial support of multiple lattice sequences | 12 January 2018, 18:59:37 UTC |
| c36448c | Vadim Mazalov | 17 December 2017, 04:34:05 UTC | Able to run for single sequence | 12 January 2018, 18:59:36 UTC |
| 71880a9 | Vadim Mazalov | 14 December 2017, 23:25:52 UTC | Able to run up to lattice consupmtion in the SE node | 12 January 2018, 18:59:36 UTC |
| 98594d5 | Vadim Mazalov | 14 December 2017, 05:22:54 UTC | Enabled more SE parameters | 12 January 2018, 18:59:36 UTC |
| 6e485ad | Vadim Mazalov | 13 December 2017, 20:11:14 UTC | Further work to enable lattice deserializer in Sequence Node | 12 January 2018, 18:57:27 UTC |
| 30bf710 | Vadim Mazalov | 13 December 2017, 06:07:58 UTC | Make the new lattice node inherit from the legacy one | 12 January 2018, 18:57:27 UTC |
| 113cdf0 | Vadim Mazalov | 12 December 2017, 16:22:39 UTC | Initial work on deserialization of the lattice | 12 January 2018, 18:57:27 UTC |
| 589aa30 | Eldar Akchurin | 11 December 2017, 15:46:24 UTC | Adding binary packing | 12 January 2018, 18:57:27 UTC |
| 8aa9342 | Vadim Mazalov | 10 December 2017, 08:26:28 UTC | Add symlistpath as a parameter to the SE node' | 12 January 2018, 18:57:26 UTC |
| b209dd0 | Vadim Mazalov | 10 December 2017, 01:40:44 UTC | Refactor lattice methods | 12 January 2018, 18:57:25 UTC |
| fe31b52 | Vadim Mazalov | 09 December 2017, 18:32:37 UTC | Ensure that lattice metadata is passed inside the node | 12 January 2018, 18:57:25 UTC |
| b2edf0c | Vadim Mazalov | 09 December 2017, 08:56:07 UTC | Rename lattice file parameters | 12 January 2018, 18:57:25 UTC |
| 5d7e995 | Vadim Mazalov | 09 December 2017, 06:35:56 UTC | Add parameters for the lattice symlit | 12 January 2018, 18:57:24 UTC |
| 4a96f60 | Vadim Mazalov | 08 December 2017, 06:27:50 UTC | Ensure lattice consumption in MB | 12 January 2018, 18:57:24 UTC |
| 1b2a256 | Vadim Mazalov | 07 December 2017, 03:05:43 UTC | Add SequenceWithLatticeNode | 12 January 2018, 18:57:23 UTC |
| 5a01480 | Vadim Mazalov | 06 December 2017, 07:18:57 UTC | Update the lattice reader sample size | 12 January 2018, 18:51:50 UTC |
| 4731e40 | Vadim Mazalov | 05 December 2017, 22:36:37 UTC | Update SGD to not consume lattice for new serializer | 12 January 2018, 18:51:50 UTC |
| b88cac1 | Vadim Mazalov | 03 December 2017, 23:08:06 UTC | Refine lattice reader changes | 12 January 2018, 18:51:50 UTC |
| f9539f8 | Vadim Mazalov | 02 December 2017, 22:18:27 UTC | make the initial version compile | 12 January 2018, 18:51:50 UTC |
| e43e23f | Vadim Mazalov | 01 December 2017, 01:39:09 UTC | Lattice reader, add lattice index | 12 January 2018, 18:51:50 UTC |
| 31c8c1e | Vadim Mazalov | 29 November 2017, 16:11:44 UTC | Lattice deserializer - index | 12 January 2018, 18:51:50 UTC |
| e1f1101 | Vadim Mazalov | 28 November 2017, 19:06:56 UTC | Lattice reader, cont | 12 January 2018, 18:51:50 UTC |
| 2758a8c | Vadim Mazalov | 25 November 2017, 01:25:58 UTC | Lattice reader, cont | 12 January 2018, 18:51:49 UTC |
| 0b76ebc | Vadim Mazalov | 23 November 2017, 13:28:24 UTC | Introduce lattice reader | 12 January 2018, 18:51:49 UTC |
